• Alves, O., , M. Balmaseda, , D. Anderson, , and T. Stockdale, 2004: Sensitivity of dynamical seasonal forecasts to ocean initial conditions. Quart. J. Roy. Meteor. Soc., 130 , 647668.

    • Search Google Scholar
    • Export Citation
  • Anderson, D. L. T., , and M. Balmaseda, 2005: Overview of ocean models at ECMWF. Proc. Seminar on Recent Developments in Numerical Methods for Atmospheric and Ocean Modelling, Reading, United Kingdom, ECMWF, 103–111.

  • Anderson, D. L. T., and Coauthors, 2003: Comparison of the ECMWF seasonal forecast systems 1 and 2, including the relative performance for the 1997/8 El Nino. ECMWF Tech. Memo. 404, 58 pp.

  • Anderson, D. L. T., and Coauthors, 2007: Development of the ECMWF seasonal forecast system 3. ECMWF Tech. Memo. 503, 95 pp.

  • Balmaseda, M. A., , A. Vidard, , D. Anderson, , A. Troccoli, , and J. Vialard, 2004: Ocean data assimilation for seasonal forecasts. Proc. Seminar on Recent Developments in Data Assimilation for Atmosphere and Ocean, Reading, United Kingdom, ECMWF, 301–326.

  • Balmaseda, M. A., 2005: Ocean analysis at ECMWF: From real-time ocean initial conditions to historical ocean reanalysis. ECMWF Newsletter, No. 105, 24–42. [Available online at http://www.ecmwf.int/publications/newsletters/pdf/105.pdf.].

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., , D. L. T. Anderson, , and F. Molteni, 2007a: Climate variability from the ECMWF system 3 ocean reanalysis. ECMWF Newsletter, No. 113, 8–16. [Available online at http://www.ecmwf.int/publications/newsletters/pdf/113.pdf.].

  • Balmaseda, M. A., , D. L. T. Anderson, , and A. Vidard, 2007b: Impact of Argo on analyses of the global ocean. Geophys. Res. Lett., 34 .L16605, doi:10.1029/2007GL030452.

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., , D. L. T. Anderson, , and A. Vidard, 2007c: The ECMWF system 3 ocean analysis. ECMWF Tech. Memo. 508, 49 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/-pdf/tm/501-600/tm508.pdf.].

  • Balmaseda, M. A., , D. Dee, , A. Vidard, , and D. L. T. Anderson, 2007d: A Multivariate treatment of bias for sequential data assimilation: Application to the tropical oceans. Quart. J. Roy. Meteor. Soc., 133 , 167179.

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., , G. C. Smith, , K. Haines, , D. Anderson, , T. N. Palmer, , and A. Vidard, 2007e: Historical reconstruction of the Atlantic Meridional Overturning Circulation from the ECMWF operational ocean reanalysis. Geophys. Res. Lett., 34 .L23615, doi:10.1029/2007GL031645.

    • Search Google Scholar
    • Export Citation
  • Bloom, S. C., , L. L. Takacs, , A. M. Da Silva, , and D. Ledvina, 1996: Data assimilation using incremental analysis updates. Mon. Wea. Rev., 124 , 12561271.

    • Search Google Scholar
    • Export Citation
  • Bonjean, F., , and G. S. E. Lagerloef, 2002: Diagnostic model and analysis of the surface currents in the tropical Pacific Ocean. J. Phys. Oceanogr., 32 , 29382954.

    • Search Google Scholar
    • Export Citation
  • Burgers, G., , M. Balmaseda, , F. Vossepoel, , G. J. van Oldenborgh, , and P. J. van Leeuwen, 2002: Balanced ocean-data assimilation near the equator. J. Phys. Oceanogr., 32 , 25092519.

    • Search Google Scholar
    • Export Citation
  • Church, J., , and N. J. White, 2006: A 20th century acceleration in global sea-level rise. Geophys. Res. Lett., 33 .L01602, doi:10.1029/2005GL024826.

    • Search Google Scholar
    • Export Citation
  • Cooper, M. C., , and K. Haines, 1996: Data assimilation with water property conservation. J. Geophys. Res., 101 , C1. 10591077.

  • Dee, D., 2005: Bias and data assimilation. Quart. J. Roy. Meteor. Soc., 131 , 33233343.

  • Ducet, N., , P-Y. Le Traon, , and G. Reverdin, 2000: Global high resolution mapping of ocean circulation from TOPEX/Poseidon and ERS-1 and -2. J. Geophys. Res., 105 , 1947719498.

    • Search Google Scholar
    • Export Citation
  • Haines, K., , J. Blower, , J-P. Drecourt, , C. Liu, , A. Vidard, , I. Astin, , and X. Zhou, 2006: Salinity assimilation using S(T): Covariance relationships. Mon. Wea. Rev., 134 , 759771.

    • Search Google Scholar
    • Export Citation
  • Ingleby, B., , and M. Huddleston, 2007: Quality control of ocean temperature and salinity profiles—Historical and real-time data. J. Mar. Syst., 65 , 158175. doi:10.1016/j.jmarsys.2005. 11.019.

    • Search Google Scholar
    • Export Citation
  • Le Traon, P-Y., , F. Nadal, , and N. Ducet, 1998: An improved mapping method of multisatellite altimeter data. J. Atmos. Oceanic Technol., 15 , 522534.

    • Search Google Scholar
    • Export Citation
  • Peters, H., , M. C. Gregg, , and J. M. Toole, 1988: On the parameterization of equatorial turbulence. J. Geophys. Res., 93 , 11991218.

  • Reynolds, R., , N. Rayner, , T. Smith, , D. Stokes, , and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15 , 16091625.

    • Search Google Scholar
    • Export Citation
  • Smith, N., , J. Blomley, , and G. Meyers, 1991: A univariate statistical interpolation scheme for subsurface thermal analyses in the tropical oceans. Prog. Oceanogr., 28 , 219256.

    • Search Google Scholar
    • Export Citation
  • Smith, T. M., , and R. W. Reynolds, 1998: A high-resolution global sea surface temperature climatology for the 1961–90 base period. J. Climate, 11 , 33203323.

    • Search Google Scholar
    • Export Citation
  • Sterl, A., , and S. Caires, 2005: Climatology, variability and extrema of ocean —The Web-based KNMI/ERA-40 wave atlas. Int. J. Climatol., 25 , 963997. doi:10.1029/joc.1175.

    • Search Google Scholar
    • Export Citation
  • Stockdale, T. N., , D. L. T. Anderson, , J. O. S. Alves, , and M. A. Balmaseda, 1998: Global seasonal rainfall forecasts using a coupled ocean–atmosphere model. Nature, 392 , 370373.

    • Search Google Scholar
    • Export Citation
  • Troccoli, A., , and K. Haines, 1999: Use of the temperature–salinity relation in a data assimilation context. J. Atmos. Oceanic Technol., 16 , 20112025.

    • Search Google Scholar
    • Export Citation
  • Troccoli, A., , and P. Kallberg, 2004: Precipitation correction in the ERA-40 reanalysis. ERA-40 Project Report Series, Vol. 13, ECMWF, Reading, United Kingdom, 10 pp.

  • Troccoli, A., , M. Balmaseda, , J. Schneider, , J. Vialard, , and D. Anderson, 2002: Salinity adjustments in the presence of temperature data assimilation. Mon. Wea. Rev., 130 , 89102.

    • Search Google Scholar
    • Export Citation
  • Uppala, S., and Coauthors, 2005: The ERA-40 reanalysis. Quart. J. Roy. Meteor. Soc., 131B , 29613012.

  • van Oldenborgh, G. J., , M. A. Balmaseda, , L. Ferranti, , T. N. Stockdale, , and D. L. T. Anderson, 2005a: Did the ECMWF seasonal forecast model outperform statistical ENSO forecast models over the last 15 years? J. Climate, 18 , 32403249.

    • Search Google Scholar
    • Export Citation
  • van Oldenborgh, G. J., , M. A. Balmaseda, , L. Ferranti, , T. N. Stockdale, , and D. L. T. Anderson, 2005b: Evaluation of atmospheric fields from the ECMWF seasonal forecasts over a 15-year period. J. Climate, 18 , 32503269. Corrigendum. 18 , 51885198.

    • Search Google Scholar
    • Export Citation
  • Vialard, J., , F. Vitart, , M. A. Balmaseda, , T. N. Stockdale, , and D. T. L. Anderson, 2005: An ensemble generation method for seasonal forecasting with an ocean-atmosphere coupled model. Mon. Wea. Rev., 131 , 13791395.

    • Search Google Scholar
    • Export Citation
  • Vidard, A., , D. L. T. Anderson, , and M. Balmaseda, 2007: Impact of ocean observation systems on ocean analysis and seasonal forecasts. Mon. Wea. Rev., 135, 409–429.

    • Search Google Scholar
    • Export Citation
  • Vidard, A., , M. Balmaseda, , and D. Anderson, 2008: Assimilation of altimeter data in the ECMWF ocean analysis system. ECMWF Tech. Memo. 546, 24 pp. [Available online at http://www.ecmwf.int/publications/library/do/references/show?id= 88462.].

  • Vitart, F., 2005: Monthly forecast and the summer 2003 heat wave over Europe: A case study. Atmos. Sci. Lett., 6 , 2. 112117.

  • Weaver, A. T., , C. Deltel, , E. Machu, , S. Ricci, , and N. Daget, 2005: A multivariate balance operator for variational ocean data assimilation. Quart. J. Roy. Meteor. Soc., 131 , 36053625.

    • Search Google Scholar
    • Export Citation
  • Wolff, J., , E. Maier-Reimer, , and S. Legutke, 1997: The Hamburg Ocean Primitive Equation model. Deutsches Klimarechenzentrum Tech. Rep. 13, Hamburg, Germany, 98 pp.

  • View in gallery

    (top) The surface forcing used in the ocean analysis and the initial conditions for the calibration hindcasts for ORA-S3. (bottom) The origin of the subsurface data and surface temperature fields used.

  • View in gallery

    Typical 10-day distribution of temperature profiles in the ORA-S3 ocean analysis. (top left) the coverage, (top right) the quality control, and (bottom left) superobbing decisions. (Source: http://www.ecmwf.int/products/forecasts/d/charts/ocean/reanalysis/obsmap)

  • View in gallery

    Temperature increment coming from the assimilation of three observations at different locations in the equatorial Pacific using the (top) ORA-S2 and (bottom) ORA-S3 covariances. The background temperature field is also plotted.

  • View in gallery

    Equatorial longitude–depth section of (a),(c) the mean temperature increment and (b),(d) the vertical velocity for ORA-S3 and ORA-S2, respectively. Contours are 2 × 10−4 °C h−1 for the temperature increment and 0.2 m day−1 for the vertical velocity. The zero contour is not plotted. The mean corresponds to the time average over the period 1987–2001.

  • View in gallery

    Vertical profiles of the mean difference analysis minus the observations (°C) in the (left) western and (right) eastern Pacific, for ORA-S3 (black) and ORA-S2 (gray). In both regions the bias in ORA-S3 is smaller than in ORA-S2.

  • View in gallery

    Correlation with TAO currents: ORA-S2 (gray) and ORA-S3 (black).

  • View in gallery

    Equatorial longitude–depth section of mean salinity increment in ORA-S3 for the (a) balanced and (b) unbalanced components. Contour interval is 0.210−5 psu h−1. The zero contour is not plotted. Values above (below) the reference level are in light (dark) shading. The mean corresponds to the time average over the period 1987–2001.

  • View in gallery

    Equatorial zonal section of the mean difference ORA-S3 minus ORA-nobs in (a) temperature and (b) salinity over the period 1987–2001. Contour interval is 0.2°C for temperature and 0.05 psu for salinity.

  • View in gallery

    Impact of data assimilation: the mean ORA-S3 minus ORA-nobs differences for (a) T300 and (c) S300. The average assimilation increment in the upper 300 m for (b) temperature and (d) salinity. The statistics are for the period 1987–2001. Contour interval is 0.2°C for T300, 0.04 psu for T300, 5 × 10−5 °C h−1 for the temperature increment, and 5 × 10−6 psu h−1 for the salinity increment.

  • View in gallery

    The equivalent wind stress corrections in the (a) zonal and (b) meridional directions resulting from vertically averaging the bias correction in the upper 60 m. For reference, also shown are the applied (c) zonal and (d) meridional components of the wind stress. Note the different color scale for the zonal bias correction. The units are 0.01 newtons per meter squared. The statistics are for the period 1987–2001.

  • View in gallery

    Mean difference between analyses and observations in the equatorial oceans. The top two rows show the differences for temperature (°C), and the bottom rows show the differences for salinity (psu). ORA-S3 is shown in black, and ORA-nobs is shown in gray.

  • View in gallery

    Correlations with the OSCAR zonal component of the surface currents for (top) ORA-S3 and (bottom) ORA-nobs. The contour interval is irregular (0.5, 0.6, 0.7, 0.8, 0.85, 0.9, 0.95, and 0.97).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 316 316 27
PDF Downloads 169 169 10

The ECMWF Ocean Analysis System: ORA-S3

View More View Less
  • 1 European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom
  • | 2 L’Institut National de Recherche en Informatique et en Automatique (INRIA), Grenoble, France
  • | 3 European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom
© Get Permissions
Full access

Abstract

A new operational ocean analysis/reanalysis system (ORA-S3) has been implemented at ECMWF. The reanalysis, started from 1 January 1959, is continuously maintained up to 11 days behind real time and is used to initialize seasonal forecasts as well as to provide a historical representation of the ocean for climate studies. It has several innovative features, including an online bias-correction algorithm, the assimilation of salinity data on temperature surfaces, and the assimilation of altimeter-derived sea level anomalies and global sea level trends. It is designed to reduce spurious climate variability in the resulting ocean reanalysis due to the nonstationary nature of the observing system, while still taking advantage of the observation information. The new analysis system is compared with the previous operational version; the equatorial temperature biases are reduced and equatorial currents are improved. The impact of assimilation in the ocean state is discussed by diagnosis of the assimilation increment and bias correction terms. The resulting analysis not only improves the fit to the data, but also improves the representation of the interannual variability. In addition to the basic analysis, a real-time analysis is produced (RT-S3). This is needed for monthly forecasts and in the future may be needed for shorter-range forecasts. It is initialized from the near-real-time ORA-S3 and run each day from it.

Corresponding author address: Magalena Alonso Balmaseda, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading RG2 9AX, United Kingdom. Email: neh@ecmwf.int

Abstract

A new operational ocean analysis/reanalysis system (ORA-S3) has been implemented at ECMWF. The reanalysis, started from 1 January 1959, is continuously maintained up to 11 days behind real time and is used to initialize seasonal forecasts as well as to provide a historical representation of the ocean for climate studies. It has several innovative features, including an online bias-correction algorithm, the assimilation of salinity data on temperature surfaces, and the assimilation of altimeter-derived sea level anomalies and global sea level trends. It is designed to reduce spurious climate variability in the resulting ocean reanalysis due to the nonstationary nature of the observing system, while still taking advantage of the observation information. The new analysis system is compared with the previous operational version; the equatorial temperature biases are reduced and equatorial currents are improved. The impact of assimilation in the ocean state is discussed by diagnosis of the assimilation increment and bias correction terms. The resulting analysis not only improves the fit to the data, but also improves the representation of the interannual variability. In addition to the basic analysis, a real-time analysis is produced (RT-S3). This is needed for monthly forecasts and in the future may be needed for shorter-range forecasts. It is initialized from the near-real-time ORA-S3 and run each day from it.

Corresponding author address: Magalena Alonso Balmaseda, European Centre for Medium-Range Weather Forecasts, Shinfield Park, Reading RG2 9AX, United Kingdom. Email: neh@ecmwf.int

1. Introduction

A new operational ocean analysis system has been implemented at the ECMWF (see Table 1 for a list of acronyms used in this paper) to provide initial conditions for seasonal forecasts. In what follows we will refer to this system as ORA-S3. The ORA-S3 ocean analysis extends back to 1959 and can be regarded as a historical ocean reanalysis that is continuously updated on a daily basis. This ocean reanalysis is used to initialize the calibrating hindcasts of the S3 seasonal forecasting system (Anderson et al. 2007) for the period 1981–2005. The earlier period of ORA-S3 ocean analysis has been used to initialize seasonal and decadal predictions within the ENSEMBLES project. As well as providing initial conditions for coupled model forecasts, the ORA-S3 ocean reanalysis, based on the synthesis of surface and subsurface ocean observations, surface fluxes from atmospheric analyses and reanalyses, and a general circulation ocean model, constitutes an important resource for climate variability studies (Balmaseda et al. 2007a, e). The ocean analysis system also has a real-time component (RT-S3), used to initialize the monthly forecasts (Vitart 2005). It is envisaged that in the near future it will also be used to initialize coupled medium-range weather forecasts.

Since 1997, ECMWF has produced operational daily global ocean analyses to provide initial conditions for the seasonal forecasting system. There have been two versions of the ocean analysis, linked to the operational seasonal forecasting system. System 1 (ORA-S1) started in 1997 (Alves et al. 2004) and provided the initial conditions for the first ECMWF operational seasonal forecasting system (Stockdale et al. 1998). System 2 (ORA-S2) was introduced in 2001 (Balmaseda 2004) and has provided initial conditions for the ECMWF operational seasonal forecasts since 2002 (Anderson et al. 2003; van Oldenborgh et al. 2005a, b; Vialard et al. 2005). A comparison between ORA-S2 and ORA-S1 ocean analyses is given in Balmaseda (2004). In 2004 an extension of ORA-S2 was introduced in order to initialize the monthly forecasting system (Balmaseda 2005). In summer 2006 the new ocean analysis system, consisting of the ORA-S3 ocean reanalysis and its real-time counterpart (RT-S3), was implemented operationally and used to initialize the seasonal forecasts from March 2007 and the monthly forecasts from June 2007. The data from the ORA-S3 are publically available online (http://ensembles.ecmwf.int/thredds/catalog.html). A selection of graphical products from both the historical reanalysis and in real time can also be found online (http://www.ecmwf.int/products/forecasts/d/charts/ocean/).

This paper describes the different components of the ORA-S3 ocean analysis system and compares its performance with the previous operational system (ORA-S2). The impact of data assimilation on the mean state and internannual variability is also discussed, with emphasis on the tropical oceans. This paper is organized as follows: section 2 offers a summary of the new features in ORA-S3 relative to ORA-S2; section 3 describes the observational datasets and the quality control procedure; section 4 provides a detailed description of the assimilation algorithm and analysis step; in section 5 the impact of assimilation is assessed by considering the assimilation increment, the bias correction term, and the fit to the data; the RT-S3 real-time extension is presented briefly in section 6.

2. Summary of new features: ORA-S3 versus ORA-S2

The ocean data assimilation system for ORA-S3 is based on the HOPE-OI scheme. The first guess is given by forcing the ocean model with daily fluxes of momentum, heat, and freshwater, while the observations are assimilated using an optimal interpolation (OI) scheme. Although the OI is no longer the most advanced assimilation method, it is a robust scheme that has proven reliable in operational use. More advanced schemes based on 3DVAR and 4DVAR are currently under development (Weaver et al. 2005).

Several modifications to the original HOPE ocean model (Wolff et al. 1997) have taken place over the years at ECMWF (Balmaseda 2004; Anderson and Balmaseda 2005): The horizontal resolution was increased to 1° × 1° with equatorial refinement, that is, the meridional resolution increases gradually toward the equator, where it is 0.3° in the meridional direction; there are 29 levels in the vertical, with a typical vertical thickness of 10 m in the upper ocean; the vertical mixing is based on Peters et al. (1988); and the barotropic solver, originally implicit, was made explicit as described in Anderson and Balmaseda (2005).

In ORA-S3, major upgrades have been introduced into the assimilation system. In addition to subsurface temperature, altimeter-derived sea level anomalies and salinity data are now assimilated. All the observations in the upper 2000 m are assimilated, whereas in ORA-S2 only the observations in the upper 400 m were used. In ORA-S3, the observations come from the quality controlled dataset prepared for the ENACT and ENSEMBLES projects until 2004 (Ingleby and Huddleston 2007), and from the Global Telecommunication System thereafter (ENACT/GTS), whereas in ORA-S2 the data were from the Global Temperature–Salinity Profile Program. More details about the data used in ORA-S3 are given in the following section. The OI scheme is now three-dimensional, the analysis being performed at all levels simultaneously down to 2000 m, whereas in ORA-S2, the analysis was carried out on each model level independently and only to 400 m. In addition, the decorrelation scales depend on the density gradient, which favors the propagation of information along isopycnals.

The first guess is obtained by integrating the ocean model from one analysis time to the next, forced by ERA-40 fluxes for the period from January 1959 to June 2002 and NWP operational analyses thereafter. In ORA-S2, the fluxes were from ERA-15/OPS, but the wind stresses were not directly used: instead, the wind stress was derived from the analyzed winds using an offline bulk formula. The representation of the upper-ocean interannual variability is improved when using the ERA-40 wind stress (Uppala et al. 2005), although the stresses are biased weak in the equatorial Pacific. The freshwater flux from ERA-40 (precipitation − evaporation, denoted PE) is known to be inaccurate. ORA-S3 uses a better estimate, obtained by “correcting” the ERA-40 precipitation values (Troccoli and Kallberg 2004). A pictorial view of the various datasets used in ORA-S3 is given in Fig. 1.

When designing a data assimilation system for seasonal forecasts, several considerations need to be taken into account. It is important to represent the interannual–decadal variability in the ocean initial conditions, and therefore strong relaxation to climatology is not desirable. On the other hand, spurious trends and signals due to the nonstationary nature of the observing system should be avoided as much as possible. It is also important to avoid large initialization shocks in the coupled model, which may damage the forecast skill. In ORA-S3, we have tried to strike a balance between the above requirements: the weight to observations has been reduced and the relaxation to climatology is considerably weaker than in ORA-S2. This has been possible because an additive bias correction has been included (Balmaseda et al. 2007d). The bias correction consists of a prescribed a priori correction to the temperature, salinity, and pressure gradient, as well as a time-dependent bias term estimated online, which acts mainly on the pressure gradient. The online bias correction is adaptive and allows for flow-dependent errors.

An important feature of the ECMWF ocean analysis system is that not just a single analysis but five simultaneous analyses are performed. The purpose of the multiple analyses is to sample uncertainty in the ocean initial conditions and to contribute to the creation of the ensemble of forecasts for the probabilistic predictions at monthly and seasonal ranges. The five simultaneous ocean analyses are created by adding perturbations to the wind stress, commensurate with the estimated uncertainty in the wind stress analysis, while the model is being integrated forward from one analysis time to the next. The wind perturbations have been revised in ORA-S3 relative to those used in ORA-S2 to better represent the perceived uncertainty in the ERA-40/OPS wind stress. Further information is available online (http://www.ecmwf.int/research/EU_projects/ENSEMBLES/documents/docu_perturbations.pdf).

3. The datasets used, observation coverage, and automatic quality control

As shown in Fig. 1, the temperature and salinity profiles prior to 1 January 2005 come from the ENACT/ENSEMBLES quality controlled dataset (Ingleby and Huddleston 2007). From January 2005, the observations come from the GTS. An automatic quality control procedure is used with several stages: first, daily averages are created if applicable as is the case for the TRITON moorings, which report hourly. Data close to the coast are rejected. A level-by-level quality control is then performed as in Smith et al. (1991), which checks both the distance between the model values and the observations in relation to the error statistics as well as the consistency between the observations by means of a buddy check. Profiles that are close in space and time are “superobbed,” following the same criteria as in Smith et al. (1991). In ORA-S3 there is an additional check for completeness of the profiles: a profile is considered incomplete, and therefore rejected, if the sparsity of the remaining observations in the vertical is judged to be insufficient to resolve the vertical temperature gradients. An observation profile will be rejected if the temperature difference between consecutive levels is larger than 5 K or if it contains a vertical temperature gradient larger than 0.1 K m−1.

Figure 2 shows a typical recent distribution of profiles in a 10-day window, in this case for the window centered around 9 January 2007. Also shown are the results of the automatic quality control decisions. Maps of the observation coverage throughout the reanalysis period can be seen online (http://www.ecmwf.int/products/forecasts/d/charts/ocean/reanalysis/obsmap). Figure 2 shows that as a result of the Argo system, the observation coverage is now almost global. Figure 2 also shows a few XBT tracks (only reporting temperature), and the TRITON/PIRATA/TAO moorings. Only the TRITON moorings in the West Pacific and Indian Oceans provide salinity in real time. The PIRATA and TAO moorings report only temperature in real time, even though the sensors measure salinity. Most of the data from XBTs and moorings are superobbed. Figure 2 shows a considerable number of Argo profiles partially rejected by our QC system in the Atlantic Ocean. Many of these partially rejected floats correspond to SOLO/FSI instruments, later reported to have faulty sensors (information online at http://www.argo.ucsd.edu).

The altimeter data consist of twice-weekly maps from AVISO. Two consecutive maps are interpolated in time to produce daily maps. Only the daily map corresponding to the center of the assimilation window is assimilated. The impact of altimeter and Argo data in the ORA-S3 system has been described in Balmaseda et al. (2007b) and Vidard et al. (2007).

The analysis of SST is not produced using the OI scheme. Instead, the model SSTs are strongly relaxed to analyzed SST, maps which are daily interpolated values derived from the OIv2 SST product (Smith and Reynolds 1998; Reynolds et al. 2002) from 1982 onward. Global SST maps from NCEP are received every Monday at midday, representing the average of the previous week’s SST values. Daily SST maps are obtained by interpolation of the weekly products. Prior to 1982, the same SST product as in the ERA-40 reanalysis was used.

4. ORA-S3 data assimilation system

In ORA-S3, the different data streams are assimilated sequentially, as illustrated by the following scheme:
i1520-0493-136-8-3018-eq1

First, the model background (b, b,η̃b, ũb) is bias corrected according to the scheme described in section 4a. Then, the detrended altimeter-derived sea level anomalies (ηo) are combined with the bias-corrected model first guess (Tb, Sb, ηb, ub) using the Cooper and Haines (1996) scheme (CH96 hereinafter) to produce a first analysis (Ta1, Sa1, ηb1, −). This analysis is then used as a first guess for a second assimilation step, where only subsurface temperature data To are assimilated, and salinity is updated by imposing conservation of the model temperature–salinity (TS) relationship, while the sea level and velocity fields remain unchanged. This second analysis is denoted as (Ta2, Sa2, −, −). In a third assimilation step, the information provided by the salinity observations So is used to modify the model TS relationship. In this step, the TS information is spread along isotherms following the scheme of Haines et al. (2006). Only salinity is modified in this step, which results in the analysis (−, Sa3, −, −). After this third assimilation step, velocity updates are derived from the temperature and salinity increments imposing geostrophic balance (Burgers et al. 2002). Finally, the trend in global (area averaged) sea level is assimilated. By combining the altimeter-derived trend in global sea level with the model trend in global dynamic height, it is possible to make the partition between the changes in the global volume and the changes in the total mass. By doing so, the global freshwater budget is closed, and the global surface salinity and sea level adjusted accordingly. Each of the steps is described further below.

The analysis is performed every 10 days. All the observations within a centered 10-day window are gathered and quality controlled. Analysis increments in temperature, salinity, and velocity are calculated using the procedure outlined above. To avoid exciting gravity waves and to allow the model dynamics to adjust gradually to the changes in the density field, an incremental analysis update (IAU) method (Bloom et al. 1996) is used: the increment is added slowly over the subsequent 10 days (IAU-10), after which a new background field is available, and the cycle is repeated.

a. Bias-correction algorithm

The presence of bias in an ocean data assimilation scheme is a serious obstacle to the reliable representation of climate by historical ocean reanalysis. The standard procedure to deal with systematic error in a data assimilation system is to augment the model state with a set of systematic error or bias variables. In sequential data assimilation, this approach requires two analysis steps: one for the bias estimation and a second for the state vector. Assuming that the bias is nearly constant in time, and that the bias error covariance matrix is proportional to the forecast error covariance matrix, with the proportionality constant small, the algorithm can be approximated so it only requires one analysis step, and thus the bias term can be updated at little extra cost (Dee 2005). However, the requirement of proportionality between the bias and forecast covariance matrices is not generally appropriate since the bias and the model state vector can have different control variables and/or multivariate balance relationships.

An alternative approximation for the one-step bias correction algorithm for the general case has been implemented in ORA-S3, following Balmaseda et al. (2007d), which uses different control variables for the bias and for the state vectors. The control variable of the bias vector is the pressure gradient, while the state vector consists of the 3D temperature, salinity, velocity, and sea level fields. The pressure gradient correction is derived from the analysis increments in the temperature. The final temperature bias is the effect of applying the pressure gradient correction to the momentum equations. Balmaseda et al. (2007d) show that applying the correction in the pressure gradient improves the equatorial circulation, whereas applying the bias correction directly to the temperature field degrades the circulation.

Modifications have also been introduced into the equation for the time evolution of the bias, which is now described by a simple parametric model as in Balmaseda et al. (2007d). The introduction of a memory term limits the influence in time of isolated or sporadic observations. A side effect is that the magnitude of the bias can be underestimated, but to compensate for that a constant term is also introduced. The constant term is not affected by the online estimation and has to be estimated a priori, preferably with independent information. The a priori term has the potential to provide a smoother analysis by preventing abrupt changes in the analysis associated with the introduction of new observing systems. In ORA-S3, the bias-correction algorithm consists of an a priori correction to the temperature, salinity, and pressure gradient, as well as a time-dependent bias estimated online. The online bias correction is only applied to the pressure gradient. The inclusion of an a priori bias-correction term enabled the subsurface relaxation to climatology to be weakened, from a time scale of 18 months in ORA-S2 to 10-yr in ORA-S3. Because of large uncertainties in the freshwater flux, the relaxation to climatology is stronger for surface salinity (∼3 yr time scale), but still considerably weaker than in the ORA-S2 analysis system, where it is ∼6 months.

b. Assimilation of altimeter-derived sea level anomalies

In ORA-S3, altimeter data are assimilated for the first time in the ECMWF operational ocean analysis. The altimeter information is given by maps of a merged satellite product, provided by SSALTO/DUACS and distributed by AVISO, with support from CNES. Twice per week (on Wednesdays and on Saturday mornings), ⅓° × ⅓° maps of sea level anomaly (MSLA) for a merged product combining available satellites, using optimal interpolation, and accounting for long-wavelength errors are produced (Le Traon et al. 1998; Ducet et al. 2000). Prior to assimilation, these maps are smoothed to remove features that are not resolved on the model grid. These are then interpolated in time to produce daily maps. Only the map corresponding to the center of the assimilation window is used.

The anomaly maps (ηo) distributed by AVISO are relative to a 7-yr mean (1 January 1993–31 December 1999). To enable comparison with the model background field (ηb), a reference mean sea level (η) is required to compute the absolute mean sea level value:
i1520-0493-136-8-3018-e1
In ORA-S3, η is the 7-yr mean sea level from an ocean analysis spanning the period 1993–99. The possibility of using a reference η derived from the GRACE gravity mission has also been explored, but it was found that sensitivities to the reference mean sea level are large, and could potentially introduce abrupt jumps in the analysis (Vidard et al. 2008).
As the ocean model uses the Boussinesq approximation, the first guess cannot represent steric changes in global volume. Therefore, prior to the assimilation, the global mean (〈ηb〉, 〈ηo〉) is removed from both the model first guess and the observations. The difference in surface elevation δη is then calculated as
i1520-0493-136-8-3018-e2
The scheme used to project sea level increment information onto subsurface temperature and salinity consists of calculating the equivalent vertical displacement of the model water column to a difference in surface elevation δη between the background and observations, subject to the constraint that the bottom pressure is not changed. Effectively, the CH96 vertical displacement of the water column translates into increments (δT1, δS1) to the temperature and salinity fields,
i1520-0493-136-8-3018-e3
from which a temporary analysis (Ta1, Sa1) is obtained:
i1520-0493-136-8-3018-e4
with ω being a weighting factor depending on latitude,
i1520-0493-136-8-3018-e5
In ORA-S3, the values for α, β, and γ are set to 0.3, 1, and 8, respectively.

c. OI assimilation of temperature

ORA-S3 uses an OI scheme for the assimilation of subsurface temperature. Originally derived from the scheme described by Smith et al. (1991), it has evolved substantially: from the original univariate two-dimensional OI scheme, where the analysis was performed on each model level separately, to a three-dimensional scheme, where the analysis is performed at all levels simultaneously, with an isopycnal formulation for the covariance matrices and a posteriori multivariate updates of velocity and salinity. The OI interpolation is carried out on overlapping subdomains of the model horizontal grid (in order to reduce the cost of the matrix inversions). Where domains overlap, the analyses are blended together. The subdivisions of the globe into subdomains depends on the observation distribution and is done so that the maximum number of observations within the domain is less than 500 (for ORA-S2, the maximum number of observations was 200).

In ORA-S3, the assimilation of subsurface temperature is performed with 3D domains down to 2000 m. In ORA-S3 there are two novel aspects: the introduction of vertical scales and the introduction of a density-dependent term, which favors an isopycnal spreading of information. The background errors BTT are represented by Gaussian functions that are anisotropic and inhomogeneous:
i1520-0493-136-8-3018-e6
The terms Wx, Wy in Eq. (6) represent the weighting given to the observations depending on the x and y distances, respectively; within 4° of the equator, the correlation length scale in the E–W direction is 1000 km while in the N–S direction it is 150 km. The correlation scales change with latitude, such that poleward of 15° it is 300 km in all directions. Between the equatorial strip and the subtropics there is a smooth transition in correlation scales. Equation (6) shows explicitly the introduction of vertical scales (Wz), which weight the influence of the observations depending on the vertical distance to the model grid point (|Δz|), as well as the introduction of a density-dependent term Wρ, which weights the observations depending on the background density difference between the observation location and the model grid point (|Δρ|), thus preventing the spread of information across water masses with very different characteristics. In ORA-S3, Wρ is 0.5 kg m−3 and Wz is depth dependent, with the values proportional to the thickness of the model levels. Observation errors are assumed to be correlated in space and time, with a spatial correlation function with length scale of 2° and a temporal correlation scale of 3 days. In general, the observations are given half the weight of the background, although the relative weight given to the data compared to the background field varies with depth to account for an increase in the uncertainty associated with the large gradients near the thermocline (see Balmaseda 2004). This function has been modified in ORA-S3 relative to ORA-S2 so that near the thermocline, the observations in ORA-S3 are given less weight than in ORA-S2 (about four times less). This reduction of the weight given to the data near the model thermocline is especially important in the equatorial Atlantic, where the model has a diffuse thermocline and the vertical stratification is weak: experimental results show that giving less weight to the observations improves the analysis in this region.

Figure 3 shows the assimilation increment coming from three single observations of temperature along the equator for an experiment using covariances as in ORA-S2 (top panel) and as in ORA-S3 (bottom panel). Contours are background temperature isotherms. One can see the effect of the two differences described above: the information is spread vertically in ORA-S3 and the spread is not uniform but depends on density. At 200°E and 210 m, where the water masses are well stratified, the increment is almost an ellipsoid, whereas at 120°W and 100 m, it has a much more complicated shape.

The assimilation of subsurface temperature follows the equation
i1520-0493-136-8-3018-e7
where the gain matrix KTT, used to calculate the increments in T, is
i1520-0493-136-8-3018-e8
Following the assimilation of temperature data, the salinity field is updated through KST. Strictly speaking, KST is not a matrix, but to simplify the notation we represent it as a gain matrix in Eq. (7). The salinity corrections are derived from the temperature increments by conserving the background water mass properties. The basic idea is to use locally the model background TS relationship to reconstruct the salinity profile from temperature information only (Troccoli and Haines 1999; Troccoli et al. 2002, hereinafter TH99). It has been shown that temperature assimilation can improve the salinity field of an ocean model by taking advantage of the large fraction of salinity variance that is strongly correlated with temperature variance. Once the Ta1 analysis has been carried out, salinity is updated following
i1520-0493-136-8-3018-e9
where δS2(z) ensures that the TS relationship is the same before and after the analysis of temperature; that is,
i1520-0493-136-8-3018-e10

The correction to S is not applied in the mixed layer or at higher latitudes where the hypothesis behind this scheme is less applicable. This is achieved by applying a latitude filter that reduces the effect of the scheme linearly to zero between 30° and 60° latitude.

d. Assimilation of salinity data on temperature surfaces

Another new feature of ORA-S3 is the assimilation of salinity data: With the recent development of the Argo network, we now have an unprecedentedly good level of spatial coverage of salinity observations.

The conventional approach to assimilating salinity is to use covariance relationships formulated in (x, y, z) coordinates. However, by doing the analysis in (x, y, z) coordinates we are not taking advantage of the fact that the salinity increments in the TH99 scheme leave the salinity unchanged on T surfaces. Haines et al. (2006) proposed an assimilation scheme by which the temperature and salinity provide two separate pieces of information about the hydrographic structure of the ocean: The temperature information is used to correct the temperature and salinity field by preserving the TS relationship, and the salinity information can be used to correct the model TS relationship. They also propose a change of variable when assimilating salinity information; instead of using geographical coordinates, the salinity assimilation is done in temperature space. We refer to this scheme as S(T) in what follows, and the conventional scheme as S(z).

Let us assume that at a particular location at level z the temperature analysis has already been performed and has yielded the temperature Ta2, and salinity analysis Sa2 by preserving the first-guess ST relationship (TH99). Following the S(T) scheme, the salinity analysis is carried out as follows:
i1520-0493-136-8-3018-e11
Written in temperature coordinates, it is clear that the correction from the S(T) scheme is correcting an entirely different aspect of the salinity field error from the correction by TH99 (Haines et al. 2006). This elegant property means that the appropriate gain KSS in the salinity assimilation and KST in the temperature assimilation are additive, which is a nice feature.
In practice we need to express the result of Eq. (11) in z coordinates. In z-level coordinates, Eq. (11) implies a second salinity increment that is additive to the first salinity increment of Eq. (9), defined as
i1520-0493-136-8-3018-e12
The δS3[Ta2(z)] salinity increment is applied in addition to the first salinity increment δS2[Ta2(z)], associated with temperature assimilation. The spreading of the information along isotherms is determined by the gain matrix KSS:
i1520-0493-136-8-3018-e13
where WRr) = exp[(−1/2)(|Δx|/Wx + |Δy|/Wy + |Δz|/Wz)] as in Eq. (8) and WT is an additional temperature scale that determines how salinity measurements on one temperature surface should influence salinity on another temperature surface. In ORA-S3, WT = 1 K. The assumption that a large fraction of salinity variance is strongly correlated with temperature variance is not valid everywhere, especially at high latitudes. For that reason the assimilation switches gradually between 30° and 60° from S(T) to S(z), using the same latitude filter as the salinity adjustment. Additionally, S(T) is not used in the mixed layer where the search for the right temperature class may fail. Sensitivity experiments show that S(T) produces a better fit to the data than S(z), not only in salinity but also in temperature. As for the temperature, the observation error RSS is diagonal, the relative ratio of the diagonal elements of BSS and RSS is the same as the ratio used for temperature, and the vertical dependence is also parameterized in the same way.

The total density increment resulting from the combined assimilation of altimeter, subsurface temperature, and salinity is used to compute velocity corrections following a geostrophic relationship (Burgers et al. 2002). The details of the implementation of the geostrophic corrections are given in Balmaseda et al. (2004).

e. Assimilation of sea level trends

The assimilation of sea level data requires careful consideration of the treatment of global sea level trends. If this trend is produced by thermal expansion due to global warming, it cannot be represented by the ocean model as, in common with most ocean models used for climate activities, the Boussinesq approximation is made, which means that ocean models preserve volume and will not see a global sea level rise in response to a global warming. Therefore, if not treated correctly, the trend in sea level can be a problem when assimilating altimeter observations. In ORA-S3, the global sea level trend is removed from the altimeter sea level anomalies before they are assimilated. The information on global sea level trends can be used, however, in combination with the trends in the ocean subsurface to determine how much of the sea level rise is due to thermal expansion and how much is due to mass changes of the ocean (Church and White 2006). In principle, ocean reanalyses provide an estimation of the steric height ηs by combining model first guesses with subsurface data. By comparing the total trend in sea level given by the altimeter data with the trend in steric height given by the ocean analysis, it is possible to estimate the component of the trend due to mass variations. In ORA-S3, the information given by the altimeter data about changes in the global mean sea level between consecutive analysis cycles (Δη) is compared every assimilation cycle with the changes in ocean analysis steric height (Δηs). The residual,
i1520-0493-136-8-3018-e14
is applied as a freshwater flux, uniform in space. The partition between volume change and mass change is valuable information, since it can help to close the freshwater budget over the oceans, although only in a global sense.

5. Results

This section focuses on the impact of data assimilation on the mean state and on the representation of ocean currents, with an emphasis on the tropics. The impact on climate variability and initialization of seasonal forecasts is shown in Balmaseda et al. 2007(a,e). A comparison with the previous operational system (ORA-S2) is provided, together with a discussion of the assimilation increment and bias correction. The impact of data assimilation is then assessed by comparing the ORA-S3 analyses with equivalent ocean analyses in which no data have been assimilated. Everything else (spinup, forcing fields, subsurface relaxation, the prescribed additive component of the bias correction, and the relaxation to SST) remains the same. In what follows we refer to this latter experiment as ORA-nobs.

a. Comparsion with ORA-S2

Figure 4 shows the equatorial section of the mean temperature increment (Fig. 4a) and the mean vertical velocity (Fig. 4b) for ORA-S3. For comparison, the corresponding figures for ORA-S2 are shown in Figs. 4c and 4d. The resulting mean increment in ORA-S3 (Fig. 4a) is smaller than in ORA-S2 (Fig. 4c), and does not show the pronounced dipolar structure. Although there are several differences between ORA-S2 and ORA-S3, Balmaseda et al. (2007c) show that the differences in the assimilation increment along the equator are mainly due to the pressure gradient bias correction, which notably reduces the spurious vertical circulation present in ORA-S2, as can be seen by comparing Figs. 4b and 4d. In addition to reducing the mean assimilation increment, the bias correction also leads to a better analysis: In ORA-S3 the fit to the data is better. Figure 5 shows the vertical profiles of the mean difference between the analyses and observations in the western (EQ3: 5°N–5°S, 150°E–170°W) and eastern Pacific (Niño-3: 5°N–5°S, 150°–90°W). Positive–negative differences are indicative of a warm–cold bias. Both the warm bias in the eastern Pacific and the cold bias in the western Pacific in ORA-S2 have been significantly reduced in ORA-S3, resulting in a stronger east–west slope of the thermocline.

The velocity measurements from the TAO moorings provide an independent dataset for the validation of ocean analyses, since the currents have not been assimilated in either ORA-S2 or ORA-S3. Figure 6 shows the vertical profiles of the time correlation of the currents in ORA-S2 (gray) and ORA-S3 (black) with the TAO currents at different mooring locations. The better correlation shown by ORA-S3 indicates that not only is the density field better constrained by the observations in ORA-S3, but it is also more dynamically consistent.

b. Impact of data assimilation

An equatorial cross section of the salinity assimilation increment in ORA-S3 is shown in Figs. 7a and 7b, separated into balanced and unbalanced components, respectively. In the eastern Pacific, the salinity increment is mainly due to the balanced operator that preserves the TS relationship (TH99): The temperature assimilation, by applying a negative temperature increment, lifts the position of the thermocline; the salinity increment from the balance operator is consistent with lifting the density profiles. The impact of the salinity observations is more noticeable in the western Pacific, where the model watermass properties are corrected by reducing the salinity.

Figure 8 shows the equatorial cross section of the differences in ORA-S3 minus ORA-nobs in both temperature (Fig. 8a) and salinity (Fig. 8b). The vertical structure of the differences suggests that in the Pacific, the main effect of the assimilation is a net increase in the heat content by deepening the thermocline. Only in the far eastern Pacific does the data assimilation have a cooling effect below the thermocline. The pattern of differences in the equatorial Indian and Atlantic Oceans suggests that the assimilation is correcting for too much vertical mixing at the base of the thermocline, consistent with the pattern of the differences in the salinity. The pattern of the difference between ORA-S3 and ORA-nobs is very different from that of the assimilation increment (shown in Fig. 4).

The nonlocal effect of the assimilation increments is also visible in Fig. 9. The left column in Fig. 9 shows the impact of data assimilation on the upper-ocean heat and salinity content, as measured by the average temperature and salinity over the upper 300 m (T300 and S300, respectively). The figures show the difference between ORA-S3 and ORA-nobs. The right column in Fig. 9 shows the average assimilation increment in the upper 300 m in temperature and salinity. By comparing the impact of the data in the left column in Fig. 9 with the applied increment in the right column it is obvious that the impact of the data is far from local. For instance, the largest temperature increment is applied in the eastern equatorial Pacific (negative), where the differences in T300 between ORA-S3 and ORA-nobs are not so obvious. Equally, the difference in ORA-S3 minus ORA-nobs in T300 in the equatorial Atlantic is quite large but the temperature increment is small. There are two reasons for the “nonlocality” of the increment: the ocean circulation, which can move the increment around, and bias correction in the pressure gradient.

In the Pacific, the data assimilation increases the equatorial ocean heat content (Fig. 9a). The heat gain has a banded latitudinal structure, with maxima on both sides of the equator (at around 2°–4°N/2°–4°S) and a larger maximum in the area of the NEC (12°–15°N). There is also a slight increase in the east–west slope of the thermal gradient. It appears as if the data assimilation is correcting for two kinds of errors in the equatorial Pacific: First, it corrects the deficit of the equatorial heat content in ORA-nobs, due to an excessive export of heat via the Indonesian Throughflow, too large meridional heat transport, or an excessive vertical mixing; second, it corrects the slope of the thermocline, which is too flat in the ORA-nobs experiment. In the equatorial Atlantic, data assimilation also steepens the east–west zonal gradient, especially by cooling the eastern part of the basin. In the tropical Indian Ocean, data assimilation decreases the heat and salinity content: the excess of heat/salt in the Indian Ocean in the ORA-nobs experiment is consistent with too large a transport via the Indonesian Throughflow. In both the tropical Pacific and Atlantic, it appears as if the whole structure associated with the subtropical gyres (both north and south of the equator) is displaced equatorward. In the northern extratropics, the differences in T300 and S300 suggest that data assimilation affects the path of the Kuroshio and Gulf Stream, which are not as zonal in ORA-S3 as in ORA-nobs. In the North Pacific, data assimilation has a large-scale impact on S300, which is much reduced in ORA-S3 relative to ORA-nobs. In the South Pacific, the salinity in ORA-S3 is higher than in ORA-nobs.

c. The bias correction

Figures 10a and 10b show the average corrections in the zonal and meridional pressure gradients (reverse sign) resulting from the online bias estimation in ORA-S3 integrated over the top 60 m. For comparison, the zonal and meridional components of the wind stress are shown in Figs. 10c,d. To make possible a detailed comparison, only the region within 10°N–10°S is shown, which is one region where the bias correction is large. The 60-m level has been chosen because below that the equatorial zonal pressure gradient is small. However, the meridional component only starts to decay below approximately 200 m. The bias correction in the pressure gradient is also large in the areas of the western boundary currents, where the correction remains significant at all depths (not shown).

In the Pacific, the equivalent correction to the zonal component is negative over most of the equator, and amounts to about 5%–10% of the mean zonal wind stress. This value is consistent with the perceived bias of the ERA-40 winds from validation of wave products (Sterl and Caires 2005). The vertical structure of the bias correction is also consistent with a wind stress bias.

In contrast, the correction to the meridional component has deeper vertical structure, larger zonal spatial correlation scale, and larger amplitude than the meridional component of the wind stress. The meridional bias term is probably related to errors in the mixing. Near the equator, east of the date line, the equivalent wind stress corresponding to the meridional pressure gradient correction shows an increased surface meridional divergence. West of the date line the meridional correction is equivalent to an increased northward flow. North of the equator, there is an increased surface meridional convergence in the area of the NECC, which is probably responsible for the increased heat content in that area. The structure of the pressure gradient correction in the Atlantic also shows an increased equatorial divergence as well as corrections for the western boundary current.

From Figs. 4, 9 and 10 it can be inferred that the increased equatorial heat content in ORA-S3 relative to ORA-nobs is likely due to the positive temperature increments in the western Pacific; the warmer water is then advected eastward by the equatorial undercurrent. The equatorial circulation in both the Pacific and Atlantic is more vigorous with data assimilation, that is, stronger equatorial undercurrents and increased divergence in the eastern part of the basins.

d. Comparison with observations

Data assimilation improves the fit to the observations, mainly by correcting the ocean mean state. Figure 11 shows the mean difference between the analyses and observations from ORA-S3 and ORA-nobs in different equatorial regions (EQ3: 5°N–5°S, 150°E–170°W; Niño-3: 5°N–5°S, 150°–90°W; EQATL: 5°N–5°S, 70°W–30°E; EQIND: 5°N–5°S, 40°–120°E). The top two rows in Fig. 11 show the mean difference in temperature and the lower two rows show the mean difference in salinity. As expected, assimilating data improves the fit to the observations. Data assimilation reduces the cold bias of ORA-nobs in the western Pacific and corrects the diffuse thermocline of the Atlantic and Indian Oceans. In general, the improvements in the mean state in both temperature and salinity extend to all geographical areas and most depths. Exceptions are the eastern Pacific (Niño-3) at thermocline depth and the North Atlantic between 50 and 150 m (not shown). In these areas the data assimilation changes the sign of the bias, with ORA-S3 being warmer than the observations while ORA-nobs is colder than the observations. For the eastern Pacific, this is a well-known problem, which was also present in ORA-S2, and which in ORA-S3 has been reduced as a result of the bias correction to the pressure gradient, but has not been completely eliminated. In general, data assimilation reduces the bias in salinity and temperature in the extratropics (not shown).

Comparison with the OSCAR currents (Bonjean and Lagerloef 2002) shows that data assimilation improves the representation of surface currents in the system, mainly in the tropical band. The Ocean Surface Current Analysis–Real-Time (OSCAR) project provides analyses of oceanic surface currents derived from satellite altimeter and scatterometer data from the end of 1992 up to near–real time and now covers the whole ocean from 60°S to 60°N. Comparisons between OSCAR and data from the Worldwide Drifter Buoy Deployment, and between OSCAR and TAO/TRITON/PIRATA mooring data, show that OSCAR products are of good quality, especially in the tropical areas (information online at http://www.oscar.noaa.gov/). Figure 12 shows the comparison with the OSCAR zonal surface velocities: these maps show the correlation between the zonal component of the surface velocities from the OSCAR monthly means and ORA-nobs and ORA-S3 for the period 1993–2006. The contour interval is nonlinear and correlations below 0.5 are not plotted. OSCAR data are not available along the coasts. Figure 12 shows that data assimilation improves the representation of the surface currents consistently: within 10° of the equator, the improvement is mainly a result of assimilating in situ temperature observations. The assimilation of sea level anomaly (SLA) improves the currents almost everywhere, though there are some exceptions such as the Gulf Stream and Kuroshio regions, and the southern oceans, where the correlation with Oscar remains below 0.5.

OSCAR currents are not completely independent from sea level data since altimetry data are used in their production. Although SLA data are used in ORA-S3, they are not assimilated directly but anomalies in T and S are derived from them and therefore there is no guarantee that the use of altimetry would lead to improved velocity analyses, and so using OSCAR currents is a useful metric to assess the quality of our analysis.

6. The real-time ocean analysis RT-S3

To make monthly forecasts, or indeed medium-range forecasts, a delay of 11 days in producing the ocean analysis is not acceptable; an ocean analysis is necessary with less than a 1-day delay. To this end, an additional ocean analysis is run each day, starting from the delayed mode analysis described above. The model is forced with the operationally analyzed surface fluxes and assimilates any available in situ data. Special treatment is required for the SST and sea level data because they are received with some delay. For the delayed ocean analysis, daily SST maps are obtained by interpolation of the weekly products, which requires the existence of two consecutive weekly values. This can introduce a delay of up to 12 days. The RT-S3 analysis does not wait for the second map of SST to become available. Instead, a daily SST product is created by adding the latest SST anomaly to the daily climatology, that is, by persisting the anomaly.

This procedure of persisting the latest anomaly is also used in the preparation of the altimeter product that goes into the real-time analysis. In the RT-S3 analysis, the daily map at the center of the assimilation window is produced either by time interpolation between two consecutive weekly maps where these are available, or by persisting the sea level anomaly from the latest map available. The anomaly is then added to the daily 1993–2001 climatology.

In the production of RT-S3, there are two assimilation cycles. The first takes place in the first time step, using all the observations in a centered 10-day window. The second is done at day 7 (5 days behind real time), also using observations in a centered 10-day window, with the first 3 days overlapping the previous cycle, and the last day (current day) with few observations. To avoid giving extra weight to the observations, only a fraction (7/10) of the assimilation increment of the first analysis cycle is applied.

7. Summary and conclusions

A new ocean analysis system has been implemented operationally at ECMWF. It consists of a historical reanalysis (ORA-S3) and a real-time component (RT-S3). The ORA-S3 reanalysis period starts in 1959 and is continuously updated with a delay of 11 days. Considerable modifications have been made to the assimilation system relative to the previous operational system (ORA-S2). An innovative bias correction algorithm has been implemented and an ocean analysis of salinity allows corrections to the watermass properties to be made. These corrections are complementary to changes to the salinity made to preserve the watermass properties in conjunction with an analysis using in situ temperature and altimetry data. Altimetric data are now used operationally but it is not yet possible to use the mean dynamic topography; only the anomalies relative to a long-term (7 yr) mean are used. Estimates of bias correction to pressure rather than temperature reduce the spurious vertical circulations that can occur in equatorial regions. The assimilation of temperature data is done in such a way as to minimize spreading across water masses with very different properties.

Relative to ORA-S2, the assimilation increment in ORA-S3 is considerably smaller, the equatorial currents are improved, and the fit to the data in the tropical oceans is better. The bias correction scheme is effective in correcting the slope of the thermocline in the equatorial Pacific. At the equator, the corrections to the zonal pressure gradient are equivalent to an increase of 5%–10% of the easterly stress, in line with estimates of the wind error derived from wave data. Corrections to the meridional component of the pressure gradient are much larger, probably related to errors in the mixing formulation in the model.

Data assimilation has a significant impact on the mean state and variability of the upper-ocean heat content. In the equatorial Pacific, it steepens the thermocline. In the Indian Ocean it sharpens the thermocline, making it shallower. The assimilation also corrects for a too diffuse thermocline in the Atlantic. The assimilation has a large impact on the salinity field. As expected, the fit to the data improves with data assimilation almost everywhere.

The assimilation scheme used in ORA-S3 has been finely tuned, and includes innovative elements such as the assimilation of salinity on temperature surfaces and bias correction schemes. However, being based on an OI-type scheme limits the potential for further development. At ECMWF, the future ocean data assimilation system will be based on the variational scheme developed by Weaver et al. (2005).

REFERENCES

  • Alves, O., , M. Balmaseda, , D. Anderson, , and T. Stockdale, 2004: Sensitivity of dynamical seasonal forecasts to ocean initial conditions. Quart. J. Roy. Meteor. Soc., 130 , 647668.

    • Search Google Scholar
    • Export Citation
  • Anderson, D. L. T., , and M. Balmaseda, 2005: Overview of ocean models at ECMWF. Proc. Seminar on Recent Developments in Numerical Methods for Atmospheric and Ocean Modelling, Reading, United Kingdom, ECMWF, 103–111.

  • Anderson, D. L. T., and Coauthors, 2003: Comparison of the ECMWF seasonal forecast systems 1 and 2, including the relative performance for the 1997/8 El Nino. ECMWF Tech. Memo. 404, 58 pp.

  • Anderson, D. L. T., and Coauthors, 2007: Development of the ECMWF seasonal forecast system 3. ECMWF Tech. Memo. 503, 95 pp.

  • Balmaseda, M. A., , A. Vidard, , D. Anderson, , A. Troccoli, , and J. Vialard, 2004: Ocean data assimilation for seasonal forecasts. Proc. Seminar on Recent Developments in Data Assimilation for Atmosphere and Ocean, Reading, United Kingdom, ECMWF, 301–326.

  • Balmaseda, M. A., 2005: Ocean analysis at ECMWF: From real-time ocean initial conditions to historical ocean reanalysis. ECMWF Newsletter, No. 105, 24–42. [Available online at http://www.ecmwf.int/publications/newsletters/pdf/105.pdf.].

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., , D. L. T. Anderson, , and F. Molteni, 2007a: Climate variability from the ECMWF system 3 ocean reanalysis. ECMWF Newsletter, No. 113, 8–16. [Available online at http://www.ecmwf.int/publications/newsletters/pdf/113.pdf.].

  • Balmaseda, M. A., , D. L. T. Anderson, , and A. Vidard, 2007b: Impact of Argo on analyses of the global ocean. Geophys. Res. Lett., 34 .L16605, doi:10.1029/2007GL030452.

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., , D. L. T. Anderson, , and A. Vidard, 2007c: The ECMWF system 3 ocean analysis. ECMWF Tech. Memo. 508, 49 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/-pdf/tm/501-600/tm508.pdf.].

  • Balmaseda, M. A., , D. Dee, , A. Vidard, , and D. L. T. Anderson, 2007d: A Multivariate treatment of bias for sequential data assimilation: Application to the tropical oceans. Quart. J. Roy. Meteor. Soc., 133 , 167179.

    • Search Google Scholar
    • Export Citation
  • Balmaseda, M. A., , G. C. Smith, , K. Haines, , D. Anderson, , T. N. Palmer, , and A. Vidard, 2007e: Historical reconstruction of the Atlantic Meridional Overturning Circulation from the ECMWF operational ocean reanalysis. Geophys. Res. Lett., 34 .L23615, doi:10.1029/2007GL031645.

    • Search Google Scholar
    • Export Citation
  • Bloom, S. C., , L. L. Takacs, , A. M. Da Silva, , and D. Ledvina, 1996: Data assimilation using incremental analysis updates. Mon. Wea. Rev., 124 , 12561271.

    • Search Google Scholar
    • Export Citation
  • Bonjean, F., , and G. S. E. Lagerloef, 2002: Diagnostic model and analysis of the surface currents in the tropical Pacific Ocean. J. Phys. Oceanogr., 32 , 29382954.

    • Search Google Scholar
    • Export Citation
  • Burgers, G., , M. Balmaseda, , F. Vossepoel, , G. J. van Oldenborgh, , and P. J. van Leeuwen, 2002: Balanced ocean-data assimilation near the equator. J. Phys. Oceanogr., 32 , 25092519.

    • Search Google Scholar
    • Export Citation
  • Church, J., , and N. J. White, 2006: A 20th century acceleration in global sea-level rise. Geophys. Res. Lett., 33 .L01602, doi:10.1029/2005GL024826.

    • Search Google Scholar
    • Export Citation
  • Cooper, M. C., , and K. Haines, 1996: Data assimilation with water property conservation. J. Geophys. Res., 101 , C1. 10591077.

  • Dee, D., 2005: Bias and data assimilation. Quart. J. Roy. Meteor. Soc., 131 , 33233343.

  • Ducet, N., , P-Y. Le Traon, , and G. Reverdin, 2000: Global high resolution mapping of ocean circulation from TOPEX/Poseidon and ERS-1 and -2. J. Geophys. Res., 105 , 1947719498.

    • Search Google Scholar
    • Export Citation
  • Haines, K., , J. Blower, , J-P. Drecourt, , C. Liu, , A. Vidard, , I. Astin, , and X. Zhou, 2006: Salinity assimilation using S(T): Covariance relationships. Mon. Wea. Rev., 134 , 759771.

    • Search Google Scholar
    • Export Citation
  • Ingleby, B., , and M. Huddleston, 2007: Quality control of ocean temperature and salinity profiles—Historical and real-time data. J. Mar. Syst., 65 , 158175. doi:10.1016/j.jmarsys.2005. 11.019.

    • Search Google Scholar
    • Export Citation
  • Le Traon, P-Y., , F. Nadal, , and N. Ducet, 1998: An improved mapping method of multisatellite altimeter data. J. Atmos. Oceanic Technol., 15 , 522534.

    • Search Google Scholar
    • Export Citation
  • Peters, H., , M. C. Gregg, , and J. M. Toole, 1988: On the parameterization of equatorial turbulence. J. Geophys. Res., 93 , 11991218.

  • Reynolds, R., , N. Rayner, , T. Smith, , D. Stokes, , and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15 , 16091625.

    • Search Google Scholar
    • Export Citation
  • Smith, N., , J. Blomley, , and G. Meyers, 1991: A univariate statistical interpolation scheme for subsurface thermal analyses in the tropical oceans. Prog. Oceanogr., 28 , 219256.

    • Search Google Scholar
    • Export Citation
  • Smith, T. M., , and R. W. Reynolds, 1998: A high-resolution global sea surface temperature climatology for the 1961–90 base period. J. Climate, 11 , 33203323.

    • Search Google Scholar
    • Export Citation
  • Sterl, A., , and S. Caires, 2005: Climatology, variability and extrema of ocean —The Web-based KNMI/ERA-40 wave atlas. Int. J. Climatol., 25 , 963997. doi:10.1029/joc.1175.

    • Search Google Scholar
    • Export Citation
  • Stockdale, T. N., , D. L. T. Anderson, , J. O. S. Alves, , and M. A. Balmaseda, 1998: Global seasonal rainfall forecasts using a coupled ocean–atmosphere model. Nature, 392 , 370373.

    • Search Google Scholar
    • Export Citation
  • Troccoli, A., , and K. Haines, 1999: Use of the temperature–salinity relation in a data assimilation context. J. Atmos. Oceanic Technol., 16 , 20112025.

    • Search Google Scholar
    • Export Citation
  • Troccoli, A., , and P. Kallberg, 2004: Precipitation correction in the ERA-40 reanalysis. ERA-40 Project Report Series, Vol. 13, ECMWF, Reading, United Kingdom, 10 pp.

  • Troccoli, A., , M. Balmaseda, , J. Schneider, , J. Vialard, , and D. Anderson, 2002: Salinity adjustments in the presence of temperature data assimilation. Mon. Wea. Rev., 130 , 89102.

    • Search Google Scholar
    • Export Citation
  • Uppala, S., and Coauthors, 2005: The ERA-40 reanalysis. Quart. J. Roy. Meteor. Soc., 131B , 29613012.

  • van Oldenborgh, G. J., , M. A. Balmaseda, , L. Ferranti, , T. N. Stockdale, , and D. L. T. Anderson, 2005a: Did the ECMWF seasonal forecast model outperform statistical ENSO forecast models over the last 15 years? J. Climate, 18 , 32403249.

    • Search Google Scholar
    • Export Citation
  • van Oldenborgh, G. J., , M. A. Balmaseda, , L. Ferranti, , T. N. Stockdale, , and D. L. T. Anderson, 2005b: Evaluation of atmospheric fields from the ECMWF seasonal forecasts over a 15-year period. J. Climate, 18 , 32503269. Corrigendum. 18 , 51885198.

    • Search Google Scholar
    • Export Citation
  • Vialard, J., , F. Vitart, , M. A. Balmaseda, , T. N. Stockdale, , and D. T. L. Anderson, 2005: An ensemble generation method for seasonal forecasting with an ocean-atmosphere coupled model. Mon. Wea. Rev., 131 , 13791395.

    • Search Google Scholar
    • Export Citation
  • Vidard, A., , D. L. T. Anderson, , and M. Balmaseda, 2007: Impact of ocean observation systems on ocean analysis and seasonal forecasts. Mon. Wea. Rev., 135, 409–429.

    • Search Google Scholar
    • Export Citation
  • Vidard, A., , M. Balmaseda, , and D. Anderson, 2008: Assimilation of altimeter data in the ECMWF ocean analysis system. ECMWF Tech. Memo. 546, 24 pp. [Available online at http://www.ecmwf.int/publications/library/do/references/show?id= 88462.].

  • Vitart, F., 2005: Monthly forecast and the summer 2003 heat wave over Europe: A case study. Atmos. Sci. Lett., 6 , 2. 112117.

  • Weaver, A. T., , C. Deltel, , E. Machu, , S. Ricci, , and N. Daget, 2005: A multivariate balance operator for variational ocean data assimilation. Quart. J. Roy. Meteor. Soc., 131 , 36053625.

    • Search Google Scholar
    • Export Citation
  • Wolff, J., , E. Maier-Reimer, , and S. Legutke, 1997: The Hamburg Ocean Primitive Equation model. Deutsches Klimarechenzentrum Tech. Rep. 13, Hamburg, Germany, 98 pp.

Fig. 1.
Fig. 1.

(top) The surface forcing used in the ocean analysis and the initial conditions for the calibration hindcasts for ORA-S3. (bottom) The origin of the subsurface data and surface temperature fields used.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 2.
Fig. 2.

Typical 10-day distribution of temperature profiles in the ORA-S3 ocean analysis. (top left) the coverage, (top right) the quality control, and (bottom left) superobbing decisions. (Source: http://www.ecmwf.int/products/forecasts/d/charts/ocean/reanalysis/obsmap)

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 3.
Fig. 3.

Temperature increment coming from the assimilation of three observations at different locations in the equatorial Pacific using the (top) ORA-S2 and (bottom) ORA-S3 covariances. The background temperature field is also plotted.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 4.
Fig. 4.

Equatorial longitude–depth section of (a),(c) the mean temperature increment and (b),(d) the vertical velocity for ORA-S3 and ORA-S2, respectively. Contours are 2 × 10−4 °C h−1 for the temperature increment and 0.2 m day−1 for the vertical velocity. The zero contour is not plotted. The mean corresponds to the time average over the period 1987–2001.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 5.
Fig. 5.

Vertical profiles of the mean difference analysis minus the observations (°C) in the (left) western and (right) eastern Pacific, for ORA-S3 (black) and ORA-S2 (gray). In both regions the bias in ORA-S3 is smaller than in ORA-S2.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 6.
Fig. 6.

Correlation with TAO currents: ORA-S2 (gray) and ORA-S3 (black).

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 7.
Fig. 7.

Equatorial longitude–depth section of mean salinity increment in ORA-S3 for the (a) balanced and (b) unbalanced components. Contour interval is 0.210−5 psu h−1. The zero contour is not plotted. Values above (below) the reference level are in light (dark) shading. The mean corresponds to the time average over the period 1987–2001.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 8.
Fig. 8.

Equatorial zonal section of the mean difference ORA-S3 minus ORA-nobs in (a) temperature and (b) salinity over the period 1987–2001. Contour interval is 0.2°C for temperature and 0.05 psu for salinity.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 9.
Fig. 9.

Impact of data assimilation: the mean ORA-S3 minus ORA-nobs differences for (a) T300 and (c) S300. The average assimilation increment in the upper 300 m for (b) temperature and (d) salinity. The statistics are for the period 1987–2001. Contour interval is 0.2°C for T300, 0.04 psu for T300, 5 × 10−5 °C h−1 for the temperature increment, and 5 × 10−6 psu h−1 for the salinity increment.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 10.
Fig. 10.

The equivalent wind stress corrections in the (a) zonal and (b) meridional directions resulting from vertically averaging the bias correction in the upper 60 m. For reference, also shown are the applied (c) zonal and (d) meridional components of the wind stress. Note the different color scale for the zonal bias correction. The units are 0.01 newtons per meter squared. The statistics are for the period 1987–2001.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 11.
Fig. 11.

Mean difference between analyses and observations in the equatorial oceans. The top two rows show the differences for temperature (°C), and the bottom rows show the differences for salinity (psu). ORA-S3 is shown in black, and ORA-nobs is shown in gray.

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Fig. 12.
Fig. 12.

Correlations with the OSCAR zonal component of the surface currents for (top) ORA-S3 and (bottom) ORA-nobs. The contour interval is irregular (0.5, 0.6, 0.7, 0.8, 0.85, 0.9, 0.95, and 0.97).

Citation: Monthly Weather Review 136, 8; 10.1175/2008MWR2433.1

Table 1.

List of acronyms used in the text.

Table 1.
Save